HYBRID ALGORITHMS FOR RELEVANT FEATURE SELECTION IN DIMENSIONALITY REDUCTION: A CASE STUDY IN DERMATOLOGICAL DISEASE CLASSIFICATION
Ключевые слова:
feature selection, dimensionality reduction, dermatological classification, hybrid algorithms, machine learningАннотация
Feature selection is a critical preprocessing step in machine learning that aims to reduce dimensionality while preserving classification performance. This study presents a comprehensive investigation of hybrid algorithms for relevant feature selection applied to dermatological disease classification. We implemented sequential forward selection with Multi-Layer Perceptron tie-breaking mechanisms to systematically identify informative features from a dermatological dataset containing 30 clinical and histopathological features. The hybrid approach achieved 95.95% accuracy using only 14 features, representing a 53.3% reduction in feature space. Key discriminators included clubbing rete ridges, vacuolization damage basal layer, and fibrosis papillary dermis. Our findings demonstrate that hybrid feature selection algorithms effectively optimize the balance between model complexity and predictive performance.
